Ochre

AI-Powered Business Development Platform for Independent Artists

Comment

Problem Statement

Independent artists — painters, illustrators, photographers, ceramicists, printmakers, and others working across visual mediums — face a structural problem that has nothing to do with talent: they are trained to make art, not to run businesses. The gap between creative skill and commercial viability is not a motivation problem or a discipline problem. It is a knowledge and sequencing problem.

Existing tools fail them in specific ways. Generic business coaching platforms like Coursera or MasterClass deliver static, linear curriculum with no awareness of where a specific artist actually is. Social media schedulers and CRM tools assume the user already has an audience and a sales process. Productivity apps like Notion or Todoist track tasks but have no opinion about which tasks matter or in what order. Artist-specific communities offer peer support but no structured progression. The result is that most independent artists cycle through bursts of effort with no compounding — they complete tasks that don't connect to each other, skip the uncomfortable ones, and plateau at the same revenue ceiling year after year.

The core problem is that there is no system that can look at a specific artist, understand where they actually are across the full spectrum of business competencies — pricing, audience building, sales conversion, platform management, brand identity, workflow — and generate a sequenced, personalized development path that adapts as they grow and confronts the specific avoidance behaviors that are holding them back.

What it does

Ochre is a web application that conducts a structured diagnostic interview with each artist, builds a detailed competency profile across 20+ business dimensions, and generates a personalized 3-phase roadmap of concrete tasks. As the artist completes tasks, checks in weekly, and progresses through roadmap cycles, the system tracks behavioral patterns — specifically what they skip and avoid — and uses that data to generate increasingly difficult, increasingly personalized continuation roadmaps that build directly on prior completed work.

The platform is not a course and not a task manager. It is a closed-loop development system: onboarding feeds the interview, the interview feeds competency scoring, competency scoring feeds roadmap generation, task completion and check-ins feed behavioral analysis, and behavioral analysis feeds the next roadmap cycle. Every component produces data that every other component consumes.

Key Features

*Adaptive Diagnostic Interview * The interview is conducted by a Gemini 2.5 Flash AI acting as a business strategist, not a chatbot. It runs across five structured sections — artistic foundation, business reality, audience and marketing, goals and vision, and blockers and readiness — with 3 to 6 questions per section. The AI tracks what has already been answered across sections and never repeats topics. Each section ends with a structured assessment of which competencies were demonstrated and what goals were extracted. At completion, a separate extraction pass pulls explicit pain points, avoidance behaviors, existing strengths, price points, audience description, primary fear or block, root cause analysis, readiness for change, and quickest win from the full conversation.

Competency Mapping with Strict Progression Rules The system tracks 20+ named business competencies — including pricing fundamentals, audience clarity, sales conversion, platform management, brand identity, and others — each across six states: hidden, unaware, early, developing, established, advanced. Progression between states is not automatic or AI-decided. It requires verified evidence: advancing from early to developing requires at minimum 4 completed tasks in that competency, 1 check-in session that mentioned it, and zero active avoidance patterns. Advancing from established to advanced requires 15 completed tasks averaging difficulty 3.5+, 6 check-in mentions, no avoidance patterns, and 90 days elapsed at the established level. Downgrade is also possible: if behavioral analysis detects 5 or more avoidance events in a competency, the system downgrades it by one level.

Personalized 3-Phase Roadmap Generation After the interview, Gemini generates a 3-phase roadmap with 7 to 9 tasks per phase. Each task has a title, step-by-step description, general rationale, personalized rationale referencing the artist's specific interview responses, estimated minutes, difficulty score on a 1–5 scale, a flag for whether it directly confronts an avoidance behavior, and an outcome evidence field — one sentence describing what the artist can physically show to prove the task is genuinely complete. Tasks within phases have increasing difficulty. Phase 1 tasks run difficulty 1–2, Phase 2 runs 2–3, Phase 3 runs 3–4. Phases 2 and 3 are locked until Phase 1 is completed.

Cycle Continuation with Compounding Difficulty When an artist completes all three phases of a roadmap cycle, the system generates a new 3-phase roadmap for the next cycle. The continuation generator fetches the full history — all previous roadmap steps, all completed and skipped tasks, all competency states, the 5 most recent check-in sessions, the interview extraction, the artist profile, capability scores, and behavioral state — and builds a structured narrative context that is fed to Gemini. The minimum difficulty floor for each new cycle is the previous cycle's average task difficulty plus 0.5, capped at 4. Every task in a continuation roadmap must reference something specific from the artist's history using phrases like "Building on your [previous task title]." The AI is required to introduce at least 2 competencies that were previously at unaware or early state.

Behavioral Analysis Agent A background agent analyzes patterns across task completions and skips to identify avoidance behaviors. If an artist repeatedly skips tasks in a specific competency — for example, consistently skipping anything related to pricing or public visibility — the system records this as an avoidance pattern with a count, updates the behavioral state, and flags future tasks in that competency as blocks avoidance behavior. These tasks are then prioritized in roadmap generation and cannot be quietly bypassed without the system registering the pattern.

Weekly Check-In System Artists complete structured weekly check-ins that capture current mood, what they worked on, what they avoided, and what felt hard. Check-in data feeds directly into competency signal resolution — a competency mentioned in a check-in session counts toward the evidence threshold required for state advancement. The check-in history is also passed to the continuation roadmap generator so the AI can reference recent emotional and practical context when writing personalized task rationales.

Weekly Report and Email Digest A weekly report agent aggregates the past 7 days of activity — tasks completed, competencies advanced, streak data, readiness delta — and generates a written summary. This is delivered via Resend email with the artist's name, progress metrics, and a forward-looking note on what to focus on next.

Dashboard with Progress Intelligence The dashboard surfaces: overall readiness score and week-over-week delta, current streak in days, tasks completed this week and all-time, the active roadmap step and its task completion percentage, competency insights showing which areas are strengths and which have avoidance patterns, the quickest win from the interview extraction, and the current roadmap narrative arc. A stage diagnosis — one of three bottleneck clusters (foundation, growth engine, or business operations) — is shown with a plain-language explanation of why the artist is stuck at their current level.

Subscription and Access Control Stripe handles subscriptions with monthly and annual pricing tiers. The middleware enforces subscription status on every protected route — dashboard, roadmap, tasks, interview, and check-in — by querying the subscriptions table directly. Stripe webhooks handle subscription creation, updates, and cancellation in real time. Users without an active subscription are redirected to the subscribe page regardless of authentication status.

Challenges we ran into

The challenge I didn't fully anticipate was behavioral design.

It's technically straightforward to detect that someone skipped a task. It's genuinely hard to design a system that responds to that avoidance productively — not by nagging, not by removing the task, but by generating a smaller, lower-friction version of the same confrontation and making it unavoidable in the next cycle. Getting that loop right, and making it feel like a coach rather than a surveillance system, is still the thing I think about most.

Accomplishments that we're proud of

The thing we're most proud of isn't a feature — it's the loop. Building a system where every user action produces data that makes the next recommendation more accurate, without any manual curation, is genuinely hard to get right. The closed feedback loop between interview, competency scoring, roadmap generation, behavioral tracking, and cycle continuation works as a coherent system, not a collection of features bolted together.

We're proud of the behavioral avoidance detection. It's easy to track what someone does. It's much harder to build a system that notices what someone keeps not doing, understands why it matters, and responds by making the avoidance unavoidable rather than optional in the next cycle — without feeling punitive. The competency progression rules are something we'd defend in any room. Rejecting the instinct to let AI decide when someone has "leveled up" — and instead requiring verified, multi-dimensional evidence across completed tasks, elapsed time, check-in mentions, and zero active avoidance — makes the system honest in a way most AI products aren't.

And we're proud that the interview actually works. Getting Gemini to behave like a structured strategist across five sections, track what's already been covered, never repeat topics, and produce a reliable extraction pass at the end — that took significant iteration and produces something genuinely useful on the first session.

What we learned

The hardest technical problem wasn't the AI. It was designing a system where everything produces data that everything else consumes. The interview feeds competency scoring. Competency scoring feeds roadmap generation. Task completions and skips feed behavioral analysis. Behavioral analysis feeds the next roadmap cycle. Getting that closed loop right — so that a task skipped six weeks ago quietly surfaces as a prioritized confrontation in a future roadmap — required rethinking how I structured every database table and every API call.

The hardest product problem was restraint. The instinct when building with AI is to make it do everything. I had to repeatedly strip features back to a single question: does this make the artist more likely to do the next right thing? If not, it was theater. The competency progression rules — requiring verified evidence across completed tasks, check-in mentions, elapsed time, and zero active avoidance before allowing state advancement — came directly from that discipline. The system can't be fooled by a burst of activity. It tracks patterns, not events.

The Gemini integration pushed me deepest. Structured JSON output at scale, across eight distinct AI use cases with different schemas and constraints, required building a discipline around prompt engineering I hadn't previously had: explicit output contracts, fallback extraction passes, schema validation before any database write, and prompt-level rules that prevent the model from generating vague verbs or generic rationales. Every task the system generates must reference something specific from the artist's history. That constraint alone took three full rewrites of the roadmap generation prompt to get right.

What's next for Ochre

The next two features on the roadmap are learning modules and automation agents — both building directly on the competency foundation already in place. Learning modules will give artists structured, practical content for each competency they're actively working on: short how-to guides, real examples, case study stories from other artists who've solved the same problem, and an AI chat panel for follow-up questions. The content will be generated by Gemini and stored statically in the database — personalization comes from which modules surface for which artist, not from regenerating content on demand for every user. This keeps costs controlled while still feeling tailored.

The automation agents feature takes the platform into genuinely new territory. Once an artist has been through the interview and is working through their roadmap, Ochre will surface agents built specifically around their active competencies and current tasks — tools that handle the repetitive, administrative work that eats creative time: metrics tracking, caption generation, outreach follow-ups. These agents live in a dedicated section of the app, organized by competency, and are scoped tightly to what that specific artist is actually doing right now. On pricing: the recommendation is to include core agents in the existing subscription rather than charging separately. Agents are the stickiest, most defensible part of the product — the moment an artist's workflows are running through Ochre, switching costs become very high. Use that to justify a higher base subscription price, and reserve a small tier of advanced or high-usage agents as an upsell. Fragmenting agent access with per-feature pricing at this stage would kill adoption before you learn what's actually valuable.

Built With

Share this project:

Updates